BDD-DETR：高效感知小目标的锂电池表面缺陷检测

doi:10.19799/j.cnki.2095-4239.2024.0591

摘要/Abstract

摘要：

针对锂电池外壳端面缺陷尺度和形状差异大而导致小目标缺陷识别困难等问题，提出BDD-DETR (battery defects detection-detection transformer)的锂电池表面缺陷检测算法。BDD-DETR架构在通用的特征提取模块和检测头模块间融入全新的模块特征感知与融合网络，通过自适应特征感知模块和特征融合路径从多个方向融合网络的深层与浅层特征，增强关键特征信息响应并抑制冗余特征，进一步提升模型多尺度特征融合能力和小目标感知能力；此外，为了减小缺陷边界框回归时的距离偏差和形状偏差，采用Shape IoU(shape intersection over union)损失函数训练网络模型。实验结果表明，在构建的锂电池端面缺陷数据集上，与Co-DETR(collaborative-detection transformer)比较，BDD-DETR平均精度提升了3.7%，小尺度目标检测精度提升了8.9%，平均召回率提升了1.1%，在锂电池的小目标缺陷检测性能上优于目前一些先进的目标检测方法。

关键词: 锂离子电池, 缺陷检测, Co-DETR, 特征感知与融合网络, Shape IoU损失

Abstract:

To address the challenges posed by the large scale and shape differences of defects on the end face of lithium battery casings, which complicate the detection of small target defects, we introduce a novel lithium battery surface defect detection algorithm based on battery defects detection-detection transformer (BDD-DETR). The BDD-DETR framework introduces a new feature perception and fusion network (FPFN) module between the general feature extraction and detection head modules. Through the adaptive feature perception module and the feature fusion path in FPFN, the deep and shallow features of this network from multiple directions are merged, the response of crucial feature information is enhanced, and redundant features are suppressed, which further improves the ability of the model to fuse multi-scale features and its capability to detect small objects. In addition, to minimize distance and shape deviations during defect bounding box regression, the shape intersection over union loss function is employed to train the network model. Experimental results indicate that on a constructed lithium battery end surface defect dataset, compared to the collaborative-detection transformer, BDD-DETR improves average precision by 3.7%, small-scale object detection precision by 8.9%, and average recall rate by 1.1%. Furthermore, BDD-DETR outperforms several advanced object detection approaches in detecting small defects in lithium batteries.

Key words: lithium-ion battery, defect detection, Co-DETR, feature perception and fusion network, Shape IoU loss

中图分类号:

TP 391.41

邢远秀, 刘颛玮, 邢玉峰, 王文波. BDD-DETR：高效感知小目标的锂电池表面缺陷检测[J]. 储能科学与技术, 2025, 14(1): 370-379.

Yuanxiu XING, Zhuanwei LIU, Yufeng XING, Wenbo WANG. BDD-DETR: An efficient algorithm for detecting small surface defects on lithium batteries[J]. Energy Storage Science and Technology, 2025, 14(1): 370-379.

图/表 15

图1

图2

表1

表2

BBD-DETR模型与其他检测器的性能比较"

Method	Backbone	$m A P 50$	$m A P s$	$m A P m$	$m A P L$	$A R$
Faster RCNN $[7]$	Res50	55.2	0.0	21.3	41.0	42.9
Cascade RCNN $[8]$	Res50	68.0	2.6	23.0	41.8	37.7
YOLOv5-L	CSPDarkNet	63.9	19.1	25.0	39.4	48.7
YOLOv7-L $[10]$	CSPDarkNet	51.2	17.0	29.6	25.3	43.5
YOLOv8-L	CSPDarkNet	45.0	17.1	18.1	29.8	37.2
Conditional DETR $[15]$	Res50	55.2	11.5	12.0	33.0	35.4
DINO $[16]$	Res50	73.8	18.5	24.5	49.8	55.6
CenterNet $[22]$	Res50	56.0	4.1	20.6	49.8	43.1
Co-DETR $[17]$	Res50	74.1	12.6	25.1	49.1	57.9
BDD-DETR(ours)	Res50	77.8	21.5	26.0	51.6	59.0

表2

表3

亚类划分前后划痕缺陷的检测精度对比"

Method	$m A P 50$
亚类划分前	65.8
亚类划分后	70.5

表3

表4

AFP模块的消融研究"

SRU+CRU	CBAM	$m A P 50$	$m A P S$	$A R$
×	×	72.7	20.2	56.5
√	×	75.9(+3.2)	21.2(+1.0)	59.0(+2.5)
×	√	74.6(+1.9)	10.6(-9.6)	54.9(-1.6)
√	√	77.8(+5.1)	21.5(+1.3)	59.0(+2.5)

表4

表5

SRU与CRU组件消融研究"

Item	$m A P 50$	$m A P S$	$A R$
*	74.6	10.6	54.9
SRU	74.1(-0.5)	20.6(+10.0)	54.1(-0.8)
CRU	75.5(+0.9)	21.3(+10.7)	53.0(-1.9)
CRU+SRU	73.2(-1.4)	17.3(+6.7)	57.7(+2.8)
SRU+CRU	77.8(+3.2)	21.5(+10.9)	59.0(+4.1)

表5

图3

图4

图5

图6

图7

图8

图9

表6

Shape IoU损失与GIoU损失对模型性能的影响"

Method	$m A P 50$	$m A P S$	$m A P M$	$m A P L$	$A R$
GIoU	72.0	12.9	27.8	54.3	56.3
Shape IoU	77.8	21.5	26.0	51.6	59.0

表6

参考文献 22

1	陈佳慧, 王飞, 危荃, 等. 锂电池安全性能无损检测技术研究进展[J]. 无损检测, 2022, 44(12): 72-75.
	CHEN J H, WANG F, WEI Q, et al. Research progress on nondestructive testing technology of lithium battery safety performance[J]. Nondestructive Testing Technologying, 2022, 44(12): 72-75.
2	孙浩然, 李雅雯, 韩有军, 等. 基于拓扑滤波与改进Canny算子的锂离子电池电极缺陷检测[J]. 储能科学与技术, 2022, 11(10): 3297-3305. DOI: 10.19799/j.cnki.2095-4239.2022.0167.
	SUN H R, LI Y W, HAN Y J, et al. Lithium-ion battery electrode defect detection based on topological filtering and improved canny operator[J]. Energy Storage Science and Technology, 2022, 11(10): 3297-3305. DOI: 10.19799/j.cnki.2095-4239. 2022.0167.
3	郭绍陶, 苑玮琦. 基于双高斯纹理滤波模板和极值点韦伯对比度的圆柱锂电池凹坑缺陷检测[J]. 电子学报, 2022, 50(3): 637-642.
	GUO S T, YUAN W Q. Pit defect detection of cylindrical lithium battery based on double Gaussian texture filtering template and extreme point weber contrast[J]. Acta Electronica Sinica, 2022, 50(3): 637-642.
4	罗东亮, 蔡雨萱, 杨子豪, 等. 工业缺陷检测深度学习方法综述[J]. 中国科学: 信息科学, 2022, 52(6): 1002-1039.
	LUO D L, CAI Y X, YANG Z H, et al. Survey on industrial defect detection with deep learning[J]. Scientia Sinica (Informationis), 2022, 52(6): 1002-1039.
5	陶志勇, 杜福廷, 任晓奎, 等. 基于T-VGG的太阳电池片缺陷检测[J]. 太阳能学报, 2022, 43(7): 145-151. DOI: 10.19912/j.0254-0096.tynxb.2020-1105.
	TAO Z Y, DU F T, REN X K, et al. Defect detection of solar cells based on t-vgg[J]. Acta Energiae Solaris Sinica, 2022, 43(7): 145-151. DOI: 10.19912/j.0254-0096.tynxb.2020-1105.
6	周颖, 毛立, 张燕, 等. 改进CNN的太阳电池缺陷识别方法研究[J]. 太阳能学报, 2020, 41(12): 69-76. DOI: 10.19912/j.0254-0096. 2020.12.010.
	ZHOU Y, MAO L, ZHANG Y, et al. Research on defect detection and classification for solar cells based on improved convolutional neural network[J]. Acta Energiae Solaris Sinica, 2020, 41(12): 69-76. DOI: 10.19912/j.0254-0096.2020.12.010.
7	GIRSHICK R. Fast R-CNN[C]//2015 IEEE International Conference on Computer Vision (ICCV). December 7-13, 2015, Santiago, Chile. IEEE, 2015: 1440-1448. DOI: 10.1109/ICCV. 2015.169.
8	CAI Z W, VASCONCELOS N. Cascade R-CNN: Delving into high quality object detection[C]//2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition. June 18-23, 2018, Salt Lake City, UT, USA. IEEE, 2018: 6154-6162. DOI: 10.1109/CVPR.2018.00644.
9	LIU W, ANGUELOV D, ERHAN D, et al. SSD: Single shot MultiBox detector[M]//Lecture Notes in Computer Science. Cham: Springer International Publishing, 2016: 21-37. DOI: 10.1007/978-3-319-46448-0_2.
10	WANG C Y, BOCHKOVSKIY A, LIAO H Y M. YOLOv7: Trainable bag-of-freebies sets new state-of-the-art for real-time object detectors[C]//2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). June 17-24, 2023, Vancouver, BC, Canada. IEEE, 2023: 7464-7475. DOI: 10.1109/CVPR52729. 2023.00721.
11	张曙文, 钟振宇, 朱大虎. 基于改进YOLOx网络的金属齿轮表面缺陷检测方法[J]. 激光与光电子学进展, 2023, 60(22): 280-290. DOI: 10.3788/LOP230469.
	ZHANG S W, ZHONG Z Y, ZHU D H. Gear surface defect detection method based on improved YOLOx network[J]. Laser & Optoelectronics Progress, 2023, 60(22): 280-290. DOI: 10.3788/LOP230469.
12	张晓晓, 邓承志, 吴朝明, 等. 基于改进YOLOv4的磁瓦缺陷检测算法[J]. 计算机科学, 2023, 50(S2): 389-395. DOI: 10.11896/jsjkx.230100100.
	ZHANG X X, DENG C Z, WU Z M, et al. Magnetic tile defect detection algorithm based on improved YOLOv₄[J]. Computer Science, 2023, 50(S2): 389-395. DOI: 10.11896/jsjkx.230100100.
13	邓光伟, 尤红权, 朱志松. 基于KCC-YOLOv5的铝型材表面缺陷检测[J]. 激光与光电子学进展, 2024, 61(4): 231-239. DOI: 10.3788/LOP230950.
	DENG G W, YOU H Q, ZHU Z S. Defect detection on aluminum profile surface based on KCC-YOLOv5[J]. Laser & Optoelectronics Progress, 2024, 61(4): 231-239. DOI: 10.3788/LOP230950.
14	CARION N, MASSA F, SYNNAEVE G, et al. End-to-end object detection with transformers[M]//Lecture Notes in Computer Science. Cham: Springer International Publishing, 2020: 213-229. DOI: 10.1007/978-3-030-58452-8_13.
15	MENG D P, CHEN X K, FAN Z J, et al. Conditional DETR for fast training convergence[C]//2021 IEEE/CVF International Conference on Computer Vision (ICCV). October 10-17, 2021, Montreal, QC, Canada. IEEE, 2021: 3631-3640. DOI: 10.1109/ICCV48922.2021.00363.
16	ZHANG H, LI F, LIU S L, et al. DINO: DETR with improved DeNoising anchor boxes for end-to-end object detection[C]//The Eleventh International Conference on Learning Representations(ICLR), 2023.
17	ZONG Z F, SONG G L, LIU Y. DETRs with collaborative hybrid assignments training[C]//2023 IEEE/CVF International Conference on Computer Vision (ICCV). October 1-6, 2023, Paris, France. IEEE, 2023: 6725-6735. DOI: 10.1109/ICCV51070.2023.00621.
18	ZHANG H, ZHANG S. Shape-IoU: More accurate metric considering bounding box shape and scale [EB/OL]. (2023-12-29)[2024-04-16]. https://arxiv.org/abs/2312.17663.
19	LIN T Y, DOLLÁR P, GIRSHICK R, et al. Feature pyramid networks for object detection[C]//2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). July 21-26, 2017, Honolulu, HI, USA. IEEE, 2017: 936-944. DOI: 10.1109/CVPR.2017.106.
20	LI J F, WEN Y, HE L H. SCConv: Spatial and channel reconstruction convolution for feature redundancy[C]//2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). June 17-24, 2023, Vancouver, BC, Canada. IEEE, 2023: 6153-6162. DOI: 10.1109/CVPR52729.2023.00596.
21	WOO S, PARK J, LEE J Y, et al. CBAM: Convolutional block attention module[M]//Lecture Notes in Computer Science. Cham: Springer International Publishing, 2018: 3-19. DOI: 10.1007/978-3-030-01234-2_1.
22	ZHOU X, WANG D, KRÄHENBÜHL P. Objects as points [EB/OL]. (2019-04-16)[2024-04-16]. https://arxiv.org/abs/1904.07850.

[1]	曹巍, 陈飞, 孔祥栋, 朱志成, 韩雪冰, 卢兰光, 郑岳久. 锂离子电池极片涂布工艺研究进展[J]. 储能科学与技术, 2025, 14(1): 90-103.
[2]	张建茹, 王其钰, 李庆浩, 张献英, 王碧童, 禹习谦, 李泓. 锂离子电池失效分析中的几种物性表征技术及其应用[J]. 储能科学与技术, 2025, 14(1): 286-309.
[3]	李建萱, 林琛, 周忠凯. 基于减平均优化算法与双向长短期记忆网络的锂离子电池健康状态估算[J]. 储能科学与技术, 2025, 14(1): 358-369.
[4]	叶石丰, 洪朝锋, 綦晓, 吴伟雄, 谭子健, 周奇, 张兆阳. 基于EEMD-GRU-NN锂离子电池表面温度预测方法研究[J]. 储能科学与技术, 2025, 14(1): 380-387.
[5]	刘通, 杨瑰婷, 毕辉, 梅悦旎, 刘硕, 宫勇吉, 罗文雷. 高能量密度与高功率密度兼顾型锂离子电池研究现状与展望[J]. 储能科学与技术, 2025, 14(1): 54-76.
[6]	李可, 朱顺兵, 陶致格, 王赫. 复合水系灭火剂抑制磷酸铁锂电池火灾实验[J]. 储能科学与技术, 2025, 14(1): 140-151.
[7]	张文婧, 肖伟, 伊亚辉, 钱利勤. 锂离子电池安全改性策略研究进展[J]. 储能科学与技术, 2025, 14(1): 104-123.
[8]	刘勇, 于怀汶, 刘大鹏, 穆勇, 王瀛洲, 张秀宇. 基于ABC-LSTM模型的锂离子电池剩余使用寿命预测[J]. 储能科学与技术, 2025, 14(1): 331-345.
[9]	陈峥, 彭月, 胡竞元, 申江卫, 肖仁鑫, 夏雪磊. 基于短期充电数据和增强鲸鱼优化算法的锂离子电池容量预测[J]. 储能科学与技术, 2025, 14(1): 319-330.
[10]	梅悦旎, 屈雯洁, 程广玉, 向永贵, 陆海燕, 邵晓丹, 张益明, 王可. 锂离子电池正极补锂技术研究进展[J]. 储能科学与技术, 2025, 14(1): 77-89.
[11]	孙中麟, 李嘉波, 田迪, 王志璇, 邢晓静. 基于COA-LSTM和VMD的锂离子电池剩余寿命预测[J]. 储能科学与技术, 2024, 13(9): 3254-3265.
[12]	陆继忠, 彭思敏, 李晓宇. 基于多特征量分析和LSTM-XGBoost模型的锂离子电池SOH估计方法[J]. 储能科学与技术, 2024, 13(9): 2972-2982.
[13]	胡雪峰, 常先雷, 刘肖肖, 徐威, 张文彬. 适用于宽温度范围的锂离子电池SOC估计方法[J]. 储能科学与技术, 2024, 13(9): 2983-2994.
[14]	沈思远, 刘亚坤, 罗栋煌, 李雨珺, 郝伟. 储能锂离子电池模组暂态过电压防护设计与电路研发[J]. 储能科学与技术, 2024, 13(9): 3277-3286.
[15]	陈媛, 章思源, 蔡宇晶, 黄小贺, 刘炎忠. 融合多项式特征扩展与CNN-Transformer模型的锂电池SOH估计[J]. 储能科学与技术, 2024, 13(9): 2995-3005.